1
2 An Extensive Program of Periodic Alternative Splicing Linked to Cell Cycle
3 Progression
4
5
6 Authors:
7 Daniel Dominguez1,2, Yi-Hsuan Tsai1,3, Robert Weatheritt 4, Yang Wang1,2,
8 Benjamin J. Blencowe * 4, Zefeng Wang* 1,5
9
10 Affiliations: 11 12 1 Department of Pharmacology, University of North Carolina at Chapel Hill, Chapel Hill, NC 13 2 Lineberger Comprehensive Cancer Center, University of North Carolina at Chapel Hill, Chapel 14 Hill, NC 15 3 Program in Bioinformatics and Computational Biology, University of North Carolina at Chapel 16 Hill, Chapel Hill, NC, 17 4 Donnelly Centre and Department of Molecular Genetics, University of Toronto, Canada 18 5 Key Lab of Computational Biology, CAS-MPG Partner Institute for Computational Biology, 19 Chinese Academy of Science, Shanghai, China 20
21 * Correspondence to: [email protected]; [email protected]
22 23 24 Abstract
25 Progression through the mitotic cell cycle requires periodic regulation of gene function at
26 the levels of transcription, translation, protein-protein interactions, post-translational
27 modification and degradation. However, the role of alternative splicing (AS) in the temporal
28 control of cell cycle is not well understood. By sequencing the human transcriptome through two
29 continuous cell cycles, we identify ~1,300 genes with cell cycle-dependent AS changes. These
30 genes are significantly enriched in functions linked to cell cycle control, yet they do not
31 significantly overlap genes subject to periodic changes in steady-state transcript levels. Many of
32 the periodically spliced genes are controlled by the SR protein kinase CLK1, whose level
33 undergoes cell cycle-dependent fluctuations via an auto-inhibitory circuit. Disruption of CLK1
34 causes pleiotropic cell cycle defects and loss of proliferation, whereas CLK1 over-expression is
35 associated with various cancers. These results thus reveal a large program of CLK1-regulated
36 periodic AS intimately associated with cell cycle control.
37 38 39 40 Impact statement 41
42 We reveal extensive periodic regulation of alternative splicing during the cell cycle in
43 genes linked to cell cycle functions, and involving an auto-inhibitory mechanism employing the
44 SR protein kinase CLK1.
2 45 Introduction 46 47 Alternative splicing (AS) is a critical step of gene regulation that greatly expands
48 proteomic diversity. Nearly all (>90%) human genes undergo AS and a substantial fraction of
49 the resulting isoforms are thought to have distinct functions (Pan et al., 2008; Wang et al., 2008).
50 AS is tightly controlled, and its mis-regulation is a common cause of human diseases (Wang and
51 Cooper, 2007). Generally, AS is regulated by cis-acting splicing regulatory elements that recruit
52 trans-acting splicing factors to promote or inhibit splicing (Matera and Wang, 2014; Wang and
53 Burge, 2008). Alterations in splicing factor expression have been observed in many cancers and
54 are thought to activate cancer-specific splicing programs that control cell cycle progression,
55 cellular proliferation and migration (David and Manley, 2010; Oltean and Bates, 2014).
56 Consistent with these findings, several splicing factors function as oncogenes or tumor
57 suppressors (Karni et al., 2007; Wang et al., 2014), and cancer-specific splicing alterations often
58 affect genes that function in cell cycle control (Tsai et al., 2015).
59 Progression through the mitotic cell cycle requires periodic regulation of gene function
60 that is primarily achieved through coordination of protein levels with specific cell cycle stages
61 (Harashima et al., 2013; Vermeulen et al., 2003). This temporal coordination enables timely
62 control of molecular events that ensure accurate chromatin duplication and daughter cell
63 segregation. Periodic gene function is conventionally thought to be achieved through stage-
64 dependent gene transcription (Bertoli et al., 2013), translation (Grabek et al., 2015), protein-
65 protein interactions (Satyanarayana and Kaldis, 2009), post-translational protein modifications,
66 and ubiquitin-dependent protein degradation (Mocciaro and Rape, 2012). Although AS is one of
67 the most widespread mechanisms involved in gene regulation, the relationship between the
68 global coordination of AS and the cell cycle has not been investigated.
3 69 Major families of splicing factors include the Serine-Arginine rich proteins (SR) proteins
70 and the heterogeneous nuclear ribonucleoproteins (hnRNPs), whose levels and activities vary
71 across cell types. SR proteins generally contain one or two RNA recognition motifs (RRMs) and
72 a domain rich in alternating Arg and Ser residues (RS domain). Generally, RRM domains confer
73 RNA binding specificity while the RS domain mediates protein-protein and protein-RNA
74 interactions to affect splicing (Long and Caceres, 2009; Zhou and Fu, 2013). Post-translational
75 modifications of SR proteins, most notably phosphorylation, modulate their splicing regulatory
76 capacity by altering protein localization, stability or activity (Gui et al., 1994; Lai et al., 2003;
77 Prasad et al., 1999; Shin and Manley, 2002). Dynamic changes in SR protein phosphorylation
78 have been detected after DNA damage (Edmond et al., 2011; Leva et al., 2012) and during the
79 cell cycle (Gui et al., 1994; Shin and Manley, 2002), suggesting that regulation of AS may have
80 important roles in cell cycle control. However, the functional consequences of SR protein
81 (de)phosphorylation during the cell cycle are largely unclear.
82 Through a global-scale analysis of the human transcriptome at single-nucleotide
83 resolution through two continuous cell cycles, we have identified widespread periodic changes in
84 AS that are coordinated with specific stages of the cell cycle. These periodic AS events belong to
85 a set of genes that is largely separate from the set of genes periodically regulated during the cell
86 cycle at the transcript level, yet the AS regulated set is significantly enriched in cell cycle-
87 associated functions. We further demonstrate that a significant fraction of the periodic AS events
88 is regulated by the SR protein kinase, CLK1, and that CLK itself is also subject to cell cycle-
89 dependent regulation. Moreover, inhibition or depletion of CLK1 causes pleiotropic defects in
90 mitosis that lead to cell death or G1/S arrest, suggesting that the temporal regulation of splicing
91 by CLK1 is critical for cell cycle progression. The discovery of periodic AS thus reveals a
4 92 widespread yet previously underappreciated mechanism for the regulation of gene function
93 during the cell cycle.
94
95 Results
96 Alternative splicing is coordinated with different cell cycle phases
97 To systematically investigate the regulation of AS during the cell cycle, we performed an
98 RNA-Seq analysis of synchronously dividing cells using a total of 2.3 billion reads generated
99 across all stages (G1, S, G2 and M) of two complete rounds of the cell cycle (Figure 1-figure
100 supplement 1A). To maximize the detection of regulated AS events, we used the complementary
101 analysis pipelines, MISO and VAST-TOOLS (Katz et al., 2010; Irimia et al., 2014). These
102 pipelines have different detection specificities and employ partially overlapping reference sets of
103 annotated AS events, and therefore afford a more comprehensive analysis when employed
104 together. Both pipelines were used to determine PSI (the percent of transcript with an exon
105 spliced in) and PIR (the percent of transcripts with an intron retained). Alternative exons
106 detected by both pipelines had highly correlated PSI values (Figure 1-figure supplement 1G; see
107 below). Consistent with previous results (Bar-Joseph et al., 2008; Whitfield et al., 2002),
108 transcripts from approximately 14.2% (1,182) of expressed genes displayed periodic differences
109 in steady-state levels between two or more cell cycle stages (see below). Remarkably, 15.6%
110 (1,293) of expressed genes also contained 1,747 periodically-regulated AS events, among a total
111 of ~40,000 detected splicing events (FDR < 2.5%; Figure 1A and Figure 1-figure supplement
112 1B, 1C, 1D).
113 Importantly, as has been observed previously for AS regulatory networks (Pan et al.,
114 2004), the majority of genes with periodic AS events did not overlap those with periodic steady-
5 115 state changes in mRNA expression. This indicates that genes with periodic changes in AS and
116 transcript levels are largely independently regulated during the cell cycle (Figure 1B). Further
117 supporting this conclusion, we did not observe a significant correlation (positive or negative)
118 between exon PSI values and mRNA expression levels for genes with both periodic expression
119 and periodic exon skipping (data not shown). A gene ontology (GO) analysis reveals that genes
120 with periodic AS, like those with periodic transcript level changes (Bar-Joseph et al., 2008;
121 Whitfield et al., 2002), are significantly enriched in cell cycle-related functional categories,
122 including M-phase, nuclear division and DNA metabolic process (Figure 1C; adjusted P <0.05
123 for all listed categories, FDR <10%) (Supplementary Table1). Similar GO enrichment results
124 were observed when removing the relatively small fraction (10%) of periodically spliced genes
125 that also display significant mRNA expression changes across the cell cycle (Figure 1C). These
126 results thus reveal that numerous genes not previously linked to the cell cycle, as well as
127 previously defined cell cycle-associated genes thought to be constantly expressed across the cell
128 cycle, are in fact subject to periodic regulation at the level of AS (Supplementary File 1 for a full
129 list).
130 Among the different classes of AS analyzed (cassette exons, alternative 5´/3´ splice sites
131 and intron retention [IR]), periodically regulated IR events were over-represented (relative to the
132 background frequency of annotated IR events) by ~2.2 fold whereas periodically regulated
133 cassette exons, represent the next most frequent periodic class of AS (P = 2.2×10-16, Fisher’s
134 exact test, Figure 1-figure supplement 1E). Quantitative RT-PCR assays across different cell
135 cycle stages validated periodic IR events detected by RNA-Seq (Figure 1D). Interestingly, one of
136 these IR events is in transcripts encoding aurora kinase B (AURKB), a critical mitotic factor
137 regulated at the levels of transcription, protein localization, phosphorylation and ubiquitination
6 138 (Carmena et al., 2012; Lens et al., 2010). The AURKB retained intron is predicted to introduce a
139 premature termination codon that elicits mRNA degradation through nonsense mediated decay,
140 and is thus expected to result in reduced levels of AURKB protein. The splicing of the retained
141 intron lags behind changes in the total AURKB mRNA expression (Figure 1E). We
142 computationally corrected levels of fully spliced, protein coding AURKB mRNA by taking into
143 account the fraction of intron-retaining (i.e. non-productive) transcripts across the cell cycle
144 stages (Figure 1E). The expression curve for corrected AURKB mRNA levels is substantially
145 different from total AURKB transcript levels, with a shifted peak coinciding with mitosis.
146 Periodically-regulated IR events detected in other genes, including those with known cell cycle
147 functions such as HMG20B and RAD52, are similarly expected to affect the cell cycle timing of
148 mRNA expression (Figure 1A, 1D). Collectively, these results provide evidence that the
149 temporal control of retained intron AS provides an important mechanism for establishing the
150 timing of expression of AURKB mRNA and protein, as well as of the timing of expression of
151 additional genes during the cell cycle.
152
153 The SR protein kinase CLK1 fluctuates during the cell cycle
154 Alternative splicing is generally regulated by the concerted action of multiple cis-
155 elements that recruit cognate splicing factors. Consistently, analysis of our RNA-seq data
156 revealed 96 RNA binding proteins (RBPs) with periodic mRNA expression, including RS
157 domain-containing factors like SRSF2, SRSF8, TRA2A and SRSF6 (Figure 2-figure supplement
158 1, 2A, Table 1). These 96 RBPs were significantly enriched in the GO term “splicing regulation”
159 (adjusted P = 10-4, Figure 2-figure supplement 2 2B), indicating that periodic AS is likely
160 controlled by multiple RBPs. Correlations between these RBPs and periodic splicing events were
7 161 also identified (Figure 2-figure supplement 1, 2C). For example, SRSF2 expression is
162 significantly correlated with the splicing pattern of a retained intron in the SRSF2 transcript.
163 Further supporting a role for these RBPs in controlling periodic splicing was the identification of
164 RNA motifs bound by a subset of periodically expressed RBPs (Figure 2-figure supplement 1,
165 2D). To further examine periodic RBP regulation during cell cycle, we measured the abundance
166 of known splicing regulatory proteins at different stages of the cell cycle by immunoblotting
167 (Figure 2A). Among the proteins analyzed, CDC-like kinase 1 (CLK1), an important regulator of
168 the Ser/Arg (SR) repeat family of splicing regulators, displayed the strongest cyclic expression
169 peaking at the G2/M phase (Figure 2A, 2B), consistent with the results of a recent mass-
170 spectrometry-based screen for cycling proteins (Ly et al., 2014). CLK1 is one of four human
171 CLK paralogs (CLK1-4) and is known to regulate AS via altering the phosphorylation status of
172 multiple SR proteins (Duncan et al., 1997; Jiang et al., 2009; Ninomiya et al., 2011; Prasad et al.,
173 1999). Notably, the levels of other detectable CLK paralogs, as well as members of another SR
174 protein kinase, SRPK1, did not change significantly at the level of RNA and/or protein during
175 the cell cycle (Figure 2B and Supplementary File 1).
176 Given that both CLK1 protein levels and known CLK1 substrates are periodically
177 expressed, we decided to further investigate the role of CLK1 in the context of cell cycle. The
178 levels of total CLK1 mRNA, as well as the levels of specific CLK1 splice variants, did not
179 change significantly during the cell cycle (Figure 2-figure supplement 2, 1A, 1B) indicating that
180 periodic expression of CLK1 is controlled at the level of protein translation and/or turnover.
181 Consistent with this, an exogenously expressed CLK1 protein displayed cell cycle-dependent
182 fluctuations similar to those observed for endogenous CLK1 protein (Figure 2B). Moreover,
183 CLK1 was rapidly degraded upon inhibition of translation by cycloheximide, and this effect was
8 184 reversed by co-treatment with the proteasome inhibitor MG132 (Figure 2-figure supplement 2
185 1C). Additionally, polyubiquitination of Flag-tagged CLK1 was detected following
186 immunoprecipitation with anti-Flag antibody from cells treated with MG132 (Figure 2–figure
187 supplement 2 1D). These data suggest that the levels of CLK1 protein are controlled by
188 ubiquitin-mediated degradation in a cell cycle-dependent manner.
189 Periodically regulated protein levels are often controlled through negative feedback
190 circuits involving auto-regulatory loops. CLK1 has been reported to auto-phosphorylate on
191 several residues (Ben-David et al., 1991). To investigate whether auto-phosphorylation of CLK1
192 affects its periodic regulation, we tested whether blocking its kinase activity affects its stability.
193 Inhibition of CLK1 kinase activity using a selective inhibitor, TG003 (Muraki et al., 2004),
194 markedly stabilizes both endogenous and exogenously expressed CLK1 proteins (Figure 2C).
195 Moreover, activity-dependent destabilization of CLK1 was observed with a wild type (WT)
196 protein, but not with a catalytically inactive (KD) mutant (Figure 2C, left panel). We further
197 observe that WT CLK1 is rapidly degraded upon cycloheximide treatment, whereas the KD
198 mutant is more stable (Figure 2–figure supplement 1C). We next tested whether CLK1 activity
199 is sufficient to trigger its own degradation by co-expressing KD CLK1 with increasing amounts
200 of WT CLK1. As expected, increasing amounts of WT CLK1 reduces levels of KD CLK1
201 (Figure 2D). Consistent with these results, WT CLK1 is more highly polyubiquitinated compared
202 to the KD mutant (Figure 2E, compare lanes 2 to 4), and treatment with TG003 reduces
203 polyubiquitination levels (Figure 2E, lane 2 vs. 3 and lane 4 vs. 5). Decreased polyubiquitination
204 of WT CLK1 is more prevalent than is apparent upon TG003 treatment, as CLK1 is stabilized by
205 TG003 inhibition and thus more total Flag-CLK1 is immunoprecipitated (Figure 2E). To further
206 examine whether this auto-feedback loop is required for changes in CLK1 protein levels during
9 207 the cell cycle, we treated synchronized cells with TG003 (or DMSO as a control) and measured
208 CLK1 protein levels. We observed that CLK1 inhibition prevents its turnover after the G2/M
209 phase for both endogenous and exogenously expressed kinases (Figure 2F and Figure 2–figure
210 supplement 1E). Taken together, these results provide strong evidence that CLK1 protein levels
211 are controlled by ubiquitin-mediated proteolysis in a cell cycle stage-specific manner, and that an
212 activity-dependent negative feedback loop is required for this periodic regulation. These results
213 further suggest that changes in the levels of CLK1 could account for many of the periodically
214 regulated AS transitions we have detected during the cell cycle.
215
216 CLK1 regulates AS events in genes with critical roles in cell cycle control
217 RNA-Seq analysis of cells treated with TG003 revealed 892 AS events (in 665 genes)
218 that significantly change after CLK1 inhibition (Figure 3A), including known CLK1-regulated
219 splicing events (e.g. exon 4 of CLK1 (Duncan et al., 1997)). It is worth noting that TG003 can
220 also inhibit CLK4 (although to a lesser extent than inhibition of CLK1). However, RNAi of
221 CLK1 is sufficient to recapitulate the phenotype of CLK1/4 inhibition (see below and Figure 4)
222 (Fedorov et al., 2011; Muraki et al., 2004). Intron retention and cassette exons are the most
223 overrepresented types of AS affected by TG003 (Figure 3A). Most (70%) of the CLK1-regulated
224 exons display increased skipping upon CLK1 inhibition, whereas 87% of CLK1-regulated
225 introns show increased retention (Figure 3-figure supplement 1B and 1C), consistent with a
226 recently reported role for CLK1 in the regulation of retained introns (Boutz et al., 2015). Of nine
227 analyzed TG003-affected AS events detected by RNA-Seq analysis, all were validated by semi-
228 quantitative RT-PCR assays (Figure 3B). These observations indicate that CLK1 inhibition
229 mainly suppresses splicing, consistent with a general requirement for phosphorylation of SR
10 230 proteins to promote splicing activity (Irimia et al., 2014; Prasad et al., 1999; Tsai et al., 2015).
231 Importantly, there is a significant overlap between genes with cell cycle periodic AS events
232 (Figure 1) and those with CLK1-regulated AS events, involving 156 genes (P = 8.5×10-10, hyper-
233 geometric test). In contrast, consistent with the results in Figure 1, we do not observe a
234 significant overlap between genes containing CLK1-regulated AS events and periodically
235 expressed genes. These results thus support a widespread and rapidly acting role for CLK1 in
236 controlling cell cycle-regulated AS. Indeed, CLK1 inhibition induces rapid (within 3-6 hours)
237 changes in AS among several analyzed cases (Figure 3-figure supplement 1D).
238 Supporting an important role for CLK1 in cell cycle progression, genes whose AS levels
239 are affected by CLK1 inhibition are significantly enriched in the GO terms cell cycle phase, M-
240 phase, DNA metabolic processes, nuclear division, DNA damage response, and cytokinesis
241 (adjusted P < 0.05 and FDR <20% for all listed GO terms; full list in Supplementary File 2). The
242 affected genes function at various stages of cell cycle including the G1/S transition (Figure 3C).
243 Mitotic processes were, however, associated with the largest number of CLK1-target genes with
244 AS changes and included examples that function in centriole duplication (CEP70, CEP120,
245 CEP290, CEP68, CDK5RAP2), metaphase and anaphase (e.g. CENPK, CENPE, CENPN), and
246 cytokinesis (e.g. SEPT2, SEPT10, ANLN) (additional examples in Figure 3C).
247 To further investigate the functional consequence of CLK1-dependent AS, we selected
248 two examples in genes that have important roles in the cell cycle: checkpoint kinase 2 (CHEK2),
249 a tumor suppressor that controls the cellular response to DNA damage and cell cycle entry
250 (Paronetto et al., 2011; Staalesen et al., 2004), and centromere-associated protein E (CENPE), a
251 kinetochore-associated motor protein that functions in chromosome alignment and segregation
252 during mitosis (Kim et al., 2008). We detected a TG003 dose-dependent increase in CHEK2
11 253 exon 9 inclusion, whereas overexpression of WT CLK1 induced exon 9 skipping, an event that
254 removes the CHEK2 kinase domain (Figure 3D). Expression of WT CLK1 in the presence of
255 TG003, or a catalytically inactive CLK1, had little to no effect on the splicing of this exon
256 (Figure 3D, bottom panels), indicating that the catalytic activity of CLK1 is essential for
257 regulating CHEK2 AS. Over-expression of the SR protein splicing regulator SRSF1, a known
258 target of CLK1 (Prasad et al., 1999), had a similar effect as over-expression of CLK1, resulting
259 in CHEK2 exon 9 skipping, whereas knockdown of SRSF1 had the opposite effect (Figure 3D,
260 right panel). Furthermore, the activation of CHEK2 requires homodimerization (Shen et al.,
261 2004), and we observe that the CHEK2 isoform lacking exon 9 still interacts with full-length
262 protein (Figure 3-figure supplement 1E), suggesting that this CLK1-regulated isoform may
263 function in a dominant-negative manner to attenuate CHEK2 activity.
264 CENPE is known to be tightly controlled at multiple levels (including transcription,
265 localization, phosphorylation and degradation), and disruption of its regulation leads to
266 pronounced mitotic defects. CENPE AS generates long and short isoforms (Supplementary File
267 2), with the predominant variant being the short isoform that lacks amino acids 1972-2068.
268 Inhibition of CLK1 rapidly shifts CENPE splicing to produce predominantly the long isoform
269 (Figure 3E), and this is accompanied by a reduction in CENPE protein levels during G2/M
270 phase, presumably due to the instability of the long isoform (Figure 3E). These data thus show
271 that CLK1 controls the AS of major cell cycle regulators, and therefore suggest that inhibition of
272 CLK1 may alter cell cycle progression.
273 To investigate this, we next performed an RNA-Seq analysis of synchronized cells at G1
274 and early G2 (when CLK1 accumulates), following inhibition of CLK1 with TG003. As a
275 specificity control, we performed a parallel RNA-Seq analysis using a structurally distinct CLK1
12 276 inhibitor, KHCB-19 (Fedorov et al., 2011). Strikingly, of 1,498 AS events that change between
277 G1 (t=0) and G2 (t=6) ~94% display reduced changes following treatment with the two drugs
278 (Figure 3F), with 65% commonly affected by both drugs, thus supporting an important role for
279 CLK1 in controlling cell-cycle dependent splicing.
280
281 CLK1 is Required for Normal Mitosis and Cell Proliferation
282 Given that CLK1 regulates the AS of many cell cycle factors (Figure 3C), we next
283 examined whether it is necessary for cell cycle progression. Knockdown of CLK1 using shRNAs
284 led to an accumulation of cells with 4N DNA content in multiple cell types, specifically HeLa,
285 H157, and A549 (Figure 4A, Figure 4-figure supplement 1 1A, 1B), and the extent of this
286 accumulation correlated with the degree of knockdown (Figure 4-figure supplement 1 1A). We
287 also observed a significant increase in multi-nucleation, a common consequence of defective
288 chromosome segregation or cytokinesis, following shRNA-knockdown or TG003 inhibition of
289 CLK1 in the treated cells (Figure 4B and Figure 4-figure supplement 1 1C). To visualize the
290 effect of CLK1 inhibition on mitosis at a single cell level, we performed time-lapse high-content
291 microscopy on live cells stably expressing a GFP-histone 2B fusion protein, to track changes in
292 chromatin. TG003-treated cells entered mitosis normally, as measured by nuclear envelope
293 breakdown, but displayed delayed or aberrant cytokinesis, typically resulting in multi-polar
294 divisions, increased time in metaphase, failure to undergo chromatin de-condensation and
295 eventual cell death (Figure 4C and Video 1-5). To further determine at what cell cycle stage
296 CLK1 activity is required, we inhibited CLK1 using TG003 at different time points after early S
297 phase release. Consistent with the imaging data, both control and TG003-treated cells entered
298 mitosis normally, as measured by 4N DNA content. However, inhibition of CLK1 before late S-
13 299 phase impaired progression through mitosis, whereas cells treated 5 hours after early S phase
300 release underwent a round of normal mitotic division, although failed to enter the next cell cycle
301 (Figure 4D and Figure 4-figure supplement 1 1D). These results suggest that the primary defects
302 caused by CLK1 inhibition occur in late S-phase and G2 phase, which is when CLK1 levels
303 normally begin to rise (Figure 2A). This conclusion is further supported by the observation that
304 CLK1-dependent AS targets, such as those detected in HMMR and CENPE, are periodically
305 expressed during cell cycle and peak during G2 and M phase (Figure 4-figure supplement 1 1E).
306 Taken together with the earlier results, these data support an important and multifaceted role for
307 CLK1 in the control of cell cycle progression through its function in the global regulation of
308 periodic AS.
309
310 CLK1 expression and CLK1-regulated AS is altered in kidney cancer
311 The importance of CLK1 for faithful progression through the cell cycle suggests it may
312 play a role the control of cell proliferation in cancer. Supporting this, shRNA knockdown or
313 chemical inhibition of CLK1 with TG003 or KHCB-19 in HeLa cells results in a near complete
314 block in cell proliferation, as measured by anchorage-dependent and -independent colony
315 formation assays (Figure 4E and Figure 4-figure supplement 1F). As mentioned above, CLK1 is
316 likely the primary target of inhibition in these experiments since RNAi of CLK1 recapitulates the
317 phenotype seen with these chemical inhibitors.
318 Using RNA-Seq data from the Cancer Genome Atlas (TCGA) (Cancer Genome Atlas
319 Network, 2013) we observe that CLK1 displays significantly higher expression in 72 kidney
320 tumors compared to matched normal tissue samples (Figure 4F, P = 10-5, Kolmogorov-Smirnov
321 test). Consistently, most CLK1-controlled AS events that are altered in tumors have expected
14 322 splicing changes (Figure 4-figure supplement 2 2A). Furthermore, patients with tumors that have
323 elevated CLK1 expression (i.e. the upper quartile of all samples) have significantly lower
324 survival rates relative to other patients in the comparison group (Figure 4G, P = 0.007). While
325 there was also an increase in CLK2, CLK3 and CLK4 mRNA expression in these tumors, CLK1
326 displayed the highest relative mRNA expression levels compared to CLK2 and CLK3 (Figure 4-
327 figure supplement 2 2B). Levels of the CLK4, which is ~80% identical to CLK1, did not
328 correlate with survival differences despite its increased levels in tumors (Figure 4-figure
329 supplement 2 2C). Consistent with this, CLK1-regulated AS events, as defined by the RNA-Seq
330 analysis in Figure 3, were also altered across multiple tumor types, including breast, colon, lung
331 and liver (Figure 4H and Figure 4-figure supplement 2 2D). These data are further consistent
332 with a multi-faceted role for CLK1 in regulating cell cycle progression, and also suggest that
333 CLK1 contributes to increased cell proliferation in cancer, at least in part through its role in
334 controlling periodic AS.
335
336 Discussion
337 Previous studies have shown that splicing and the cell cycle are intimately connected
338 processes. Indeed, cell cycle division (CDC) loci originally defined in S. cerevisiae, namely cdc5
339 and cdc40, were subsequently shown to encode spliceosomal components (Ben-Yehuda et al.,
340 2000; McDonald et al., 1999). Moreover, genome-wide RNAi screens for new AS regulators of
341 apoptosis genes in human cells revealed that factors involved in cell-cycle control, in addition to
342 RNA processing components, were among the most significantly enriched hits (Moore et al.,
343 2010; Tejedor et al., 2015). An RNAi screen performed in Drosophila cells for genes required
344 for cell-cycle progression identified numerous splicing components(Bjorklund et al., 2006) as
15 345 well as a Drosophila ortholog of CLK kinases, Darkener of apricot Doa (Bettencourt-Dias et al.,
346 2004). In other studies, negative control of splicing during M phase was shown to be dependent
347 on dephosphorylation of the SR family protein, SRSF10 (Shin and Manley, 2002), and the
348 mitotic regulator aurora kinase A (AURKA) was shown to control the AS regulatory activity of
349 SRSF1(Moore et al., 2010). Our study shows for the first time that AS patterns are subject to
350 extensive periodic regulation, in part via a global control mechanism involving cell cycle
351 fluctuations of the SR protein kinase CLK1. At least one likely function of this periodic AS
352 regulation is to control the timing of activation of AURKB (Figure 1), as well as of numerous
353 other key cell cycle factors shown here to be subject to periodic AS. The definition of an
354 extensive, periodically-regulated AS program in the present study thus opens the door to
355 understanding the functions of an additional layer of regulation associated with cell cycle control
356 and cancer.
357 Since many RBPs are found to be periodically expressed (Fig. 4A), it is likely that other
358 aspects of mRNA metabolism are also coordinated with cell cycle stages. For example,
359 differential degradation could potentially contribute to the observed periodic fluctuation of splice
360 isoforms during cell cycle. The degradation of mRNA is closely linked with alternative
361 polyadenylation, which has emerged as a critical mechanism that controls mRNA translation and
362 stability. Generally, shortened 3′ UTRs are found in rapidly dividing cells and more aggressive
363 cancers (Mayr and Bartel, 2009). This study identified 94 cases periodic alternative poly-A site
364 usage (data not shown) in genes known to regulate cell cycle and/or proliferation, including
365 SON, CENPF and EPCAM (Ahn et al., 2011; Bomont et al., 2005; Chaves-Perez et al., 2013). In
366 addition, many mRNAs have recently been found to be translated in a cell cycle dependent
367 fashion (Aviner et al., 2015; Maslon et al., 2014; Stumpf et al., 2013). Interestingly, a fraction of
16 368 periodically translated genes are also periodically spliced, including key regulators of cell cycle
369 (e.g., AURKA, AURKB, TTBK1 and DICER1). This observation is consistent with recent findings
370 that the regulation of AS and translation may be coupled (Sterne-Weiler et al., 2013). In
371 summary, we have demonstrated that AS is subject to extensive temporal regulation during the
372 cell cycle in a manner that appears to be highly integrated with orthogonal layers of cell cycle
373 control. These results thus provide a new perspective on cell cycle regulation that should be
374 taken into consideration when studying this fundamental biological process, both in the context
375 of normal physiology and diseases including cancers.
376
17 377 Methods
378 Cell culture and synchronizations: HeLa (a kind gift from J. Trejo), HEK 293T (from ATCC
379 CRL-3216) and A549 (kind gift from W. Kim) cells were maintained in DMEM (Gibco)
380 medium supplemented with 10% FBS (Gibco). All cells were cultured in humidified incubators
381 with 5% CO2. Cell cycle synchronization was adapted from the protocol of Whitfield et al.
382 (Whitfield et al., 2002); ~ 750,000 log phase HeLa cells were plated in 15 cm dishes in complete
383 media and allowed to attach for 16 hrs, reaching < 30% confluence. Cells were subsequently
384 treated with 2 mM thymidine (Sigma) for a total of 18 hrs, washed 2 times with 1xPBS, and
385 supplemented with fresh complete media for 10 hrs. 2 mM thymidine was subsequently added
386 for a second block of 18 hrs and washed as described previously. Mitotic block was performed
387 by double thymidine arrest (as above) and release in fresh media for 3.5 hrs followed by addition
388 of nocodazole 100 μM (Sigma) for 10 hrs. G1 block was performed by serum starvation for 72
389 hrs in DMEM containing 0.05% FBS. For RNA-Seq, cells (both adherent and detached) were
390 harvested every 1.5 hrs for 30 hrs and frozen immediately for purification of total RNA. To
391 block the activity of CLK1, cells were treated with TG003 (Sigma), KHCB-19 (Tocris). To
392 block activity of the proteasome cells were treated with MG132 (Sigma). Drugs were-suspended
393 in DMSO and added to growing cultures at the indicated concentrations and times.
394 Flow cytometry and cell cycle analysis: Cells were harvested with trypsin treatment, washed 2
395 times in cold 1xPBS and subsequently fixed in 80% ice cold ethanol for at least 4 hrs. Cells were
396 then washed twice with 1xPBS and suspended in propidium iodide/RNase staining buffer (BD
397 Pharmingen, cat # 550825). Cells were analyzed by flow cytometry to count 10,000 cells that
398 satisfied gating criteria. Data collected were analyzed using ModFit software to discern 2N (G1),
399 S-phase, and 4N (G2 and M) composition.
18 400 Mapping and filtering of RNA-Seq data: RNA-Seq reads were mapped to the human genome
401 (build hg19) using the MapSplice informatics tool with default parameters (Wang et al., 2010).
402 The mapped reads were further analyzed with Cufflinks to calculate the level of gene expression
403 with FPKM (Fragments Per Kilobase per Million mapped reads) (Trapnell et al., 2010). The
404 levels of alternatively spliced isoforms were quantified with MISO (Mixture-of-Isoforms)
405 probabilistic framework (Katz et al., 2010) using the annotated AS events for human
406 hg19version 2. The levels of alternatively spliced isoforms were also quantified with VAST-
407 TOOLS using the event annotation as previously described (Irimia et al., 2014). Each AS event
408 was assigned a PSI or PIR value to represent the percent of transcripts with the exon spliced in,
409 or the intron retained, respectively.
410 Identification of periodic AS:
411 For identification of periodic AS raw PSI/PIR values were normalized as: